System and method for rendering advertisements using lookalike model

Abstract

A system for targeting advertising comprises a processor configured to compute a similarity score between a plurality of customers in a seed set and each of a pool of in-store purchasers having no history of online purchases from the online store of an entity. Each user in the seed set has a history of online purchases from the online store of the entity and a history of in-store purchases from stores of the entity. The processor is configured for selecting a subset of the pool of in-store purchasers having similarity scores above a predetermined threshold. The processor is configured for initiating rendering of advertisements for the online store to the subset of the pool of in-store purchasers.

What is claimed is: 
     
       1. A system, comprising:
 a memory having instructions stored thereon, and a processor configured to read the instructions to: 
 filter, by the processor, customer records in a population set to generate a pool of customer records having no purchase records associated with online purchases from an online store; 
 compute a similarity score between a plurality of customer records in a seed set and the filtered pool of customer records having no purchase records associated with online purchases from the online store, wherein each customer record in the seed set includes at least one purchase record associated with online purchases and at least one purchase record associated with in-store purchases, wherein the similarity score is computed using a matching times similarity approach, wherein the matching times similarity approach considers a minimum match of a number of items for each item in each purchase record associated with online purchases normalized to a maximum match for each item in each purchase record associated with online purchases, wherein the similarity score is calculated by a trained similarity engine, and wherein the similarity engine is trained using a training set of customer records each including at least one purchase record associated with online purchases and at least one purchase record associated with in-store purchases;
 select, based on the matching times similarity approach, a subset of the filtered pool of customer records having similarity scores above a predetermined threshold; and 
 initiate rendering of a webpage including content related to the online store to at least one device associated with the subset of the filtered pool of customer records. 
 
 
     
     
       2. The system of  claim 1 , wherein computing of the similarity score is based on purchase records associated with in-store purchases of the customer records in the seed set and purchase records associated with in-store purchases of the customer records in the filtered pool. 
     
     
       3. The system of  2 , wherein the similarity score is calculated using Jaccard similarity. 
     
     
       4. The system of  3 , wherein the Jaccard similarity is calculated for each of the customer records in the filtered pool of customer records based on a number of items included in purchase records associated with the filtered pool of customer records, without regard to a quantity of each item purchased. 
     
     
       5. The system of  3 , wherein the Jaccard similarity is calculated for each of the customer records in the filtered pool of customer records based on a number of items included in purchase records associated with the filtered pool of customer records and a quantity of each item purchased. 
     
     
       6. The system of  claim 2 , wherein the similarity score is calculated using local sensitive hashing or a k-d tree technique. 
     
     
       7. The system of  1 , wherein the processor filtering of customer records in the population set to generate the filtered pool of customer records having no purchase records associated with online purchases from the online store further comprises filtering such that each customer record in the filtered pool includes at least one purchase record including at least one item from a selected class of items. 
     
     
       8. The system of  1 , wherein the processor filtering of customer records in the population set to generate the filtered pool of customer records having no purchase records associated with online purchases from the online store further comprises filtering such that each customer record in the filtered pool is associated with a predetermined demographic group. 
     
     
       9. A method, comprising:
 training, by a processor, a computer-implemented look-alike model with a training sample including a first plurality of customer records, each of the first plurality of customer records having purchase records associated with online purchases from an online store and purchase records associated with in-store purchases, wherein the computer-implemented look-alike model is configured to implement a matching times similarity approach, wherein the matching times similarity approach considers a minimum match of a number of items for each item in each purchase record associated with online purchases normalized to a maximum match for each item in each purchase record associated with online purchases; 
 filtering, by the processor, customer records in a population set to generate the pool of customer records having no purchase records associated with online purchases from an online store; 
 using, by the processor, the trained look-alike model to compute a similarity score between a seed set including a second plurality of customers and each customer record within the filtered pool of customer records having no purchase records associated with online purchases from the online store, wherein each of the second plurality of customer records includes purchase records associated with online purchases and purchase records associated with in-store purchases; 
 selecting, by the processor, a subset of the filtered pool of customer records having similarity score above a predetermined threshold; and 
 initiating, by the processor, rendering of a webpage including content related to the online store to at least one device associated with the subset of the filtered pool of customer records. 
 
     
     
       10. The method of  claim 9 , wherein the look-alike model includes a computer-implemented logistic regression model. 
     
     
       11. The method of  claim 9 , wherein the similarity score is computed based on purchase records related to in-store purchases of the customer records in the seed set and purchase records related to in-store purchases of the customer records in the filtered pool. 
     
     
       12. The method of  11 , wherein the similarity score is computed using Jaccard similarity. 
     
     
       13. The method of  12 , wherein the Jaccard similarity each of the customer records in the filtered pool is computed based on a number of items included in purchase records associated with the filtered pool of customer records, without regard to a quantity of each item purchased. 
     
     
       14. The method of  12 , wherein the Jaccard similarity each of the customer records in the filtered pool is computed based on a number of items included in purchase records associated with the filtered pool of customer records and a quantity of each item purchased. 
     
     
       15. The method of  claim 12 , wherein the similarity score is computed using local sensitive hashing or a k-d tree technique. 
     
     
       16. The method of  9 , wherein filtering, by the processor, customer records in a population set to generate the filtered pool of customer records having no purchase records associated with online purchases from the online store further comprises filtering such that each customer record in the filtered pool includes at least one purchase record including at least one item from a selected class of items. 
     
     
       17. The method of  9 , wherein filtering, by the processor, customer records in a population set to generate the filtered pool of customer records having no purchase records associated with online purchases from the online store further comprises filtering such that each customer record in the filtered pool is associated with a predetermined demographic group. 
     
     
       18. A non-transitory, machine readable storage medium encoded with program instructions, such that when the program instructions are executed by a processor, the processor performs a method, comprising:
 training a computer-implemented look-alike model with a training sample including a first plurality of customer records, each of the first plurality of customer records having purchase records associated with online purchases from an online store and purchase records associated with in-store purchases, wherein the computer-implemented look-alike model is configured to implement a matching times similarity approach, wherein the matching times similarity approach considers a minimum match of a number of items for each item in each purchase record associated with online purchases normalized to a maximum match for each item in each purchase record associated with online purchases; 
 filtering, by the processor, customer records in a population set to generate a pool of customer records having no purchase records associated with online purchases from the online store; 
 using the trained look-alike model to compute a similarity score between a seed set including a second plurality of customers and each customer record within the filtered pool of customer records having no purchase records associated with online purchases from the online store, wherein each of the second plurality of customer records includes purchase records associated with online purchases and purchase records associated with in-store purchases; 
 selecting a subset of the filtered pool of customer records having similarity score above a predetermined threshold; and 
 initiating rendering of a webpage including content related to the online store to at least one device associated with the subset of the filtered pool of customer records. 
 
     
     
       19. The method of  18 , wherein the similarity score is computed using Jaccard similarity, and the Jaccard similarity of each the customer records in the filtered pool is computed based on a number of items included in purchase records associated with the filtered pool of customer records and a quantity of each item purchased. 
     
     
       20. The system of  18 , wherein the processor filtering of customer records in a population set to generate the filtered pool of customer records having no purchase records associated with online purchases from the online store further comprises filtering such that each customer record in the filtered pool includes at least one purchase record including at least one item from a selected class of items.